Chilling Injury Segmentation of Tomato Leaves Based on Fluorescence Images and Improved k-Means++ Clustering

Highlights Chlorophyll fluorescence imaging can be used to evaluate chilling injury. Chilling injury area heterogeneity in the L*a*b* color space is significant. Improved k-means++ clustering has a good segmentation effect on chilling injury. Abstract. The application of fluorescence imaging in the detection of tomato chilling injury was investigated. With the segmentation of the chilling injury area serving as the experimental target, an algorithm based on chlorophyll fluorescence image analysis and improved k-means++ clustering was proposed. First, the extraction of lateral heterogeneity values algorithm was used to analyze the horizontal heterogeneity in five color spaces of the fluorescence images of tomato seedling leaves, and it was found that the chilling injury area was significant in the L*a*b* color space. Second, the fluorescence image was converted from the RGB color space to the L*a*b* color space, and the k-means++ algorithm was used to cluster the two-dimensional data of the a*b* space. Third, insertion sorting was used to reorder the different label regions obtained by the k-means++ clustering algorithm, and the region with the largest value was used as the target region. Finally, the binary image of the target region was filtered using a morphological noise filter, and the cold-damaged area was outputted by the mask operation. The results showed that the cold-damaged area was well segmented when the fluorescence imaging contained yellow cold traces. The mean match rate of the proposed algorithm was 37.08%, 13.52%, and 0.96% higher than that based on the HSV model and watershed algorithm, the fuzzy C-means clustering method, and the k-means clustering method, respectively. Similarly, the mean error rate was 13.69%, 5.56%, and 0.16% lower than that based on the HSV model and watershed algorithm, the fuzzy C-means clustering method, and the k-means clustering method, respectively. These findings provide a foundation for research on early warning of chilling injury by identifying the chilling injury status of tomato leaves using a computer vision method.